Gaurav ChaudharyAI & Full Stack Engineer
Blog Résumé

AI & Full Stack Engineer Bengaluru, India

I build LLM applications and real-time SaaS end to end: model routing and retrieval, the APIs and databases behind them, and the interface people use. Trained as an electronics engineer, so I measure before I guess.

Explore the work Get in touch Résumé (PDF)

Open to full-time roles and freelance builds

  • Meas 1100+businesses on Tokins
  • Meas 210K+Lastone downloads
  • Meas 3IEEEAIIoT 2024 author
Scroll to sweep

01AboutCH1 · XY 1:1

I studied electronics at VIT Vellore, wrote automotive ECU software at KPIT and evaluated large language models at Innodata. Alongside, I build and run my own products. The thread through all of it: take a noisy signal and make it dependable.

Portrait of Gaurav Chaudhary
Gaurav Chaudhary · rider_infinity on GitHub and Play

Most of my recent work sits where LLMs meet production software: routing requests across model providers, validating what comes back, and wrapping it in products that businesses use every day. Tokins, my live queue platform, serves 100+ businesses. Chills turns a job description into a tailored resume and a recruiter email.

Before that, I spent a year at KPIT on virtual ECUs for an automotive program: dSPACE tooling, timing analysis, and Python automation for setup, test runs and log collection.

Based in
Bengaluru, India
Education
B.Tech ECE, VIT Vellore, 2025
Focus
LLM apps, RAG, real-time SaaS
Published
IEEE AIIoT 2024
Certified
AWS Cloud Practitioner
Languages
English, Hindi

02ExperienceYT · 2023 → now

From storefronts to ECUs to LLMs.

Full timeline & certifications

  1. →

    InsideFPV

    Web Developer · Internship · Vellore

    • Optimised data handling and rendering on the production storefront, cutting page load times by 40%.
    • Wrote server-side logic and REST APIs in Node.js and Express, and tuned MongoDB queries.
    • Integrated secure payment gateways.

    Node.js · Express · MongoDB · AWS EC2/S3 · WordPress

  2. →

    KuppiSmart Solutions (Livestockify)

    Web Development Head · Internship · Vellore

    • Led the integration of real-time IoT sensor data into a live livestock monitoring dashboard.
    • Designed the database schema for streaming sensor data, built for low-latency live queries.
    • Built visualisations for temperature, humidity, air quality and ventilation.

    Python · ThingSpeak API · SQL · Power BI

  3. →

    KPIT Technologies

    Associate Software Engineer · Pune

    • Developed and validated ECU software for the EVCU2 program: ECU virtualisation (vECU), system integration and timing analysis.
    • Ran virtual ECU workflows on dSPACE SystemDesk, ControlDesk and VEOS Player, so validation no longer waited on hardware-in-the-loop rigs.
    • Built Python automation for environment setup, run orchestration and log collection.

    vECU · AUTOSAR · C · Python · dSPACE

  4. →

    Innodata

    Senior Associate · AI Managed Services · Noida

    • Evaluated LLM responses across task suites for accuracy, reasoning, factual consistency and instruction following.
    • Assessed RAG workflows for retrieval relevance, grounding and hallucination rate.
    • Owned AI QA: reproduced model failures, wrote evaluation rubrics and validated fixes. Automated evaluation and data validation in Python.

    LLM evaluation · RAG · Prompt engineering · Python · Pandas

  5. → Now

    AI training platforms

    Independent contractor · Remote

    • Grade and fix code written by large language models on platforms such as Handshake AI, Alignerr, Outlier and Appen.
    • Coding-evaluation tasks with prompts, rubrics and Dockerised environments, across Python, React and Vue.
    • It's where I learned exactly where LLMs break, which shapes how I build with them.

    LLM evaluation · Code review · Rubrics · Docker · Python · React

03StackCH3 · XY 3:4

From silicon to screen.

I started at the bottom of the stack, with circuits, microcontrollers and ECUs, and worked my way up to the model and the interface. This is what I reach for at each layer.

  1. L1

    Silicon & embedded

    Where I started. Virtual ECUs for an automotive program at KPIT, edge vision on an ESP32-CAM for an IEEE paper, and a RISC-V RV32I core in Verilog during a VLSI internship.

    C · Embedded C · ESP32 · STM32 · AUTOSAR · dSPACE SystemDesk, ControlDesk, VEOS · vECU · Verilog

  2. L2

    Infrastructure

    Where things have to stay up. Tokins runs on AWS; most side projects ship through Vercel or Render with containers where they help.

    AWS (EC2, S3) · Docker · CI/CD · Linux · Git & GitHub · Kubernetes · Vercel · Render

  3. L3

    Data

    Schemas for streaming IoT data, Redis in front of MongoDB for live queues, and vector indexes for retrieval.

    MongoDB · Mongoose · Redis · MySQL · Postgres · Firebase · FAISS & vector DBs · Pandas · Power BI

  4. L4

    Services

    APIs and real-time backends: queues over WebSockets, payment webhooks, auth, and model calls with retries and backoff.

    Node.js · Express · Next.js · FastAPI · REST · Socket.io · Spring Boot · GraphQL · JWT & OAuth2

  5. L5

    Intelligence

    Model routing, retrieval and evaluation. I have graded LLM output for a living, so I build with the failure modes in mind.

    LLM evaluation · RAG · Prompt engineering · Embeddings · LangChain · LangGraph · CrewAI · AutoGen · Hugging Face · Transformers · OpenCV

  6. L6

    Interface

    The part people touch: admin dashboards, an Android app with 10K+ downloads, a Chrome extension, and this site.

    React · Redux · TypeScript · HTML & CSS · Tailwind · Android (Java) · GSAP · Responsive UI

Languages Python · JavaScript · TypeScript · C/C++ · Java · SQL

04Selected workCH4 · Captures

Four builds that show how I work. Live links where they exist, source where it's public, and nothing presented as a screenshot that isn't one.

012025Founder & full-stack engineer

Tokins

A live queue and AI admin platform for clinics, pharmacies, retail and service centres.

Customers scan a QR code, get a token and follow their place in line live instead of standing around. Businesses get an admin panel with analytics, Razorpay payments and automated marketing over email, Telegram, Instagram and voice alerts, driven by an LLM pipeline.

Inside queue engine · reservations · staff & stock · multi-branch orgs · AI-written email campaigns · Instagram analytics · Telegram bot · spreadsheet import · k6 load tests

Queue-Management2024 · first idea WaitWise2025 · QR + Socket.io Tokins2025 → live

Businesses served
100+
Shorter average wait
40%
User database in load tests
1M+

React · Node.js · MongoDB · Redis · Socket.io · AWS · Razorpay · Firebase Auth · LLM

Visit tokins.in Codebase is private

Screenshot of the Tokins landing page: 'Scan. Get a Tokin. Be notified.' with a live ETA, alerts and queue status panel
Screenshot · tokins.in landing page

022026Solo build

Chills

Tailor your resume to any job, approve every change, then email the recruiter with it attached.

The resume is a LaTeX file and it stays the source of truth. The model may only rewrite the summary, skills and bullet points. Employers, dates and project headings are frozen, so it can't invent a job. Every edit is shown as a diff for approval and spliced into the exact byte range of its macro. An allow-list sanitizer runs last, because a job description is untrusted input and \input or \write18 would mean file or shell access at compile time.

Next.js · TypeScript · Postgres (Neon + Drizzle) · Claude & Gemini APIs · Zod · Nodemailer · Chrome extension

Visit chills.pro Source on GitHub

Schematic, not a screenshot: how an edit moves through Chills. Frozen macros are never touched, approved edits are spliced by byte range, and a rewrite carrying \write18 is rejected by the sanitizer.
Illustration, not footage: the camera, detector and parking-occupancy idea from the paper.

032024IEEE AIIoT 2024

Smart Surveillance

Edge object detection on an ESP32-CAM, for surveillance and parking occupancy.

OpenCV deep-learning detection tuned to fit the tight compute budget of a low-cost ESP32-CAM module, analysing live video at up to 30 FPS. Published and presented at the 3rd International Conference on AI for IoT (AIIoT 2024) with Shreya Pathak, Vikas Choudhary, Love Chauhan and Abhishek N Tripathi.

Python · OpenCV · Deep learning · ESP32-CAM · IoT

Read the paper on IEEE Xplore

042023Published as Rider Infinity

Lastone

An Android app that puts previous-year exam papers in one place, with 10K+ downloads on Google Play.

Built so that every VIT student could find past CAT and FAT papers by course code. An admin web panel uploads papers to AWS S3 and syncs the links through Firebase, so new papers show up in the app right away.

Admin panelAWS S3FirebaseAndroid app

Java · Android · Gradle · Firebase Auth · AWS S3 · Glide · Lottie

Google Play Admin panel

LabInteractiveCH3 · Chills v1 · ai.js

Watch an LLM request survive a bad day.

The first version of Chills (Express and MongoDB) sends every prompt through one function, generateContent(). It routes to the configured provider, rejects empty or refusing answers, backs off on rate limits, and repairs truncated JSON before giving up. Pick a provider, inject a fault and run it.

server/services/ai.js · generateContent() Simulation of the real code path · timings compressed 10× · nothing leaves your browser
Provider
Inject a fault
Request path through generateContent A prompt goes to the router, which picks one of five providers. The reply passes a quality gate that can trigger a retry with backoff, then JSON parsing with a repair step. promptresume text routerby config openrouter/free gemini-2.5-flash Llama-3.3-70B gpt-oss-120b claude-sonnet qualitygate JSONparse retry · backoff

Console

  1. Pick a provider and a fault, then send a request.

Read ai.js on GitHub

RIRider Infinity · independent ventureSeparate from my career · CH5 · Market

Rider Infinity. Ride the infinity edge.

My own venture, run independently of my employment. It builds and operates its own products, open-sources its trading research, and takes on a few client builds at a time with one rule: you see working software before any invoice.

  1. TKNSLive

    Tokins

    Live queue and AI admin SaaS for clinics, pharmacies and retail. Used by 100+ businesses.

    Case studytokins.in

  2. CHLLLive

    Chills

    Tailors a LaTeX resume to any job, every edit approved in a diff, then emails the recruiter.

    Case studychills.pro

  3. RIFXOpen source

    Trading engine

    Automated Forex research and execution on MetaTrader 5, built to distrust its own backtests.

    Inside the engineGitHub

  4. RI·LABTaking projects

    Client builds

    AI agents, RAG over your documents, SaaS MVPs and workflow automation. A working demo in 1 to 3 days, then you decide.

    riderinfinity.com

  5. YOUR·CONext listing

    Your company could be next.

    Send the problem in a few lines. You get a working demo of the core feature in days, before you pay anything. If it isn't right, you owe nothing.

    Start with a demo

RIFX · Rider Infinity trading engine · Python · MetaTrader 5

A trading engine that distrusts its own backtests.

Fourteen strategy modules feed an evaluation matrix. A five-gate risk manager sits between every signal and the market. Research runs on AWS Spot fleets, and a configuration only counts as an edge if it is profitable in all three independent test periods. The code even says so: one strategy's notes warn that the pattern flatters itself on hand-picked charts.

Live loop

  1. Market dataMT5 ticks & candles, auto-reconnect
  2. 14 strategiesEMA, RSI, MACD, Bollinger, regime-switch, liquidity sweep, DSL…
  3. Evaluation matrixweighted · majority · unanimous
  4. Risk manager5 gates before any order
  5. Order managerSL/TP, 5 trailing-stop profiles
  6. Journal & dashboardSQLite, Flask REST, CSV/JSON export

News intelligence Finnhub, NewsAPI and RSS → AI sentiment → time-decayed signal into the matrix

Research loop

  1. Backtesterbar-by-bar on MT5 history
  2. Optimizergrid search, ranked by profit factor or Sharpe
  3. AWS Spot fleetS3 + SQS jobs, self-terminating workers, cost cap
  4. Walk-forwardprofitable in 3 of 3 periods, or it's not an edge
RIFX · risk desk Simulated random-walk market · shows the engine's decision path · not trading results
Equity
$10,000
Today
0.0%
Trades
0
Blocked by risk
0
Evaluation matrix

Defaults are the engine's own: 1% risk, 5% daily cap, 3 positions, 30-point spread filter, ATR stops (1.5×ATR, 8 to 60 pips) with take-profit at 1.8× the stop.

  1. Press Run a session. Then try 3% risk with a 2% cap and watch the automation pause trading.
  • Regime switchingADX plus a variance-ratio test decide trend or range; in between, it stands aside.
  • Rules in plain textBUY WHEN RSI(14) < 30 AND EMA(12) > EMA(26)
  • AutomationDrawdown, volatility-spike and P&L triggers can pause trading or close every position.
  • SessionsKnows when Sydney, Tokyo, London and New York are open, and their overlaps.
  • Multi-botSeveral bots in parallel threads, each with its own symbol, strategy and timeframe.
  • Event busPub/sub backbone: market data, signals, orders and risk alerts on named channels.

Read the code Research and education only. No performance claims.

+More buildsArchive · 20+ projects

Full project archive
  • 2026

    RAG recommendation agent

    A conversational agent that recommends assessments from a 32-page catalogue it scrapes itself. Request path: pre-LLM guardrails → MiniLM embeddings and FAISS retrieval → Gemini Flash → post-LLM validation, behind a FastAPI chat endpoint.

    Python · FAISS · Gemini · FastAPI Code
  • 2026

    Duplicate image scanner

    An offline Windows app that finds images that are really the same picture, even renamed, resized, re-saved in another format or flipped. Shows each match side by side, exports to a spreadsheet, and never deletes anything.

    Python · Pillow · Windows Code
  • 2026

    SyncWave Audio

    Windows app that plays one system-audio stream on several Bluetooth or wired devices at once, with adaptive buffering and per-device latency controls.

    C# · .NET 8 WPF · WASAPI · NAudio Code
  • 2025

    Text emotion classifier

    Detects the emotion in a piece of text and replies with advice: a Keras classifier in a Python backend, plus a Node service calling Hugging Face inference models.

    Python · Keras · Node · Hugging Face Code
  • 2025

    5G resource allocation model

    Predicts spectrum and power allocation from bandwidth, latency, power, SNR and interference: a small Keras network trained on generated network data, with a front end.

    Python · TensorFlow · scikit-learn Code
  • 2024

    Indian-language speech synthesis

    Trained a VITS text-to-speech model for a non-Hindi Indian language on AI4Bharat data, on Colab GPUs, with a reproducible setup guide.

    PyTorch · VITS · TTS Code
  • 2025

    CareMate

    Django healthcare app with AI diet recommendations, ML disease prediction, dashboards and WhatsApp integration.

    Django · Python · ML LiveCode
  • 2024

    Exam paper distribution

    The admin side of Lastone: uploads exam papers to AWS S3 and syncs them to the Android app through Firebase.

    React · AWS S3 · Firebase Live
  • 2024

    Train seat allocation

    MERN app for seat booking with real-time availability and allocation logic.

    MongoDB · Express · React · Node LiveCode

LOGNotesCH2 · newest first

Writing it down is how I debug my own thinking.

All posts

Notes are published from the CMS.Open the blog

05Education & credentialsCH1 · XY 3:2

8.36/ 10 CGPA

B.Tech, Electronics and Communication Engineering

Vellore Institute of Technology, Vellore · →

Coursework: Data Structures & Algorithms · DBMS · Operating Systems · OOP

IEEE · AIIoT 2024

Smart Surveillance Object Detection System

2024 3rd International Conference on Artificial Intelligence for Internet of Things. Presented in May 2024; Certificate of Presentation from IEEE Madras Section.

IEEE Xplore
  • AWS Certified Cloud PractitionerAmazon Web Services
  • Mastering Data Structures & Algorithms using C and C++Udemy
  • Embedded Systems Bare-Metal Programming (STM32)Udemy
  • VLSI Design Internship: RISC-V RV32I RTL in VerilogMaven Silicon
  • EntrepreneurshipNPTEL
  • Developing Applications with SQL, Databases, and DjangoIBM · verify

FFTSide channelsSpectrum view · off the clock

What else shows up on the spectrum.

  • MarketsTrades stocks and Forex. RIFX grew out of it.
  • ChessA regular on Chess.com.
  • FootballPlayed through VIT.
  • SpaceExploration, an interest since my VIT days.
  • Competitive programmingOn Codeforces as gaurav_111.
  • TreesJoined the KPIT × 14 Trees plantation drive at Pargaon, 2025.
  • SustainabilityPresented lithium mining and SDG 7 with Team Infinity, 2024.
  • Bug reportsI file them for the tools I use, from Cursor to LinkedIn.
  • LanguagesHindi, English, a little Spanish.
  • HomeFrom Aligarh, building in Bengaluru.

06ContactCH1 · XY 1:2 · signal locked

Have something worth building?

I'm open to full-time AI and full-stack roles, and I take on freelance builds. Email is the fastest way to reach me.

gaurav.chaudhary.865022@gmail.com

Or leave a message